Modular Neural Network for Image Denoising Without Ground Truth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-resolution imaging techniques, such as FIB-SEM, produce noisy images due to optical and thermal effects, making it difficult to obtain noise-free images, especially at nanometer scales, and existing machine learning models require ground truth images for training which are often unavailable.
Innovation Solution
A machine learning model is trained using sequences of noisy images to identify and remove noise without ground truth images, utilizing neural networks with branched architectures and noise attention blocks to denoise images based on temporal and spatial differences within sequences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution imaging techniques are used, then image resolution is improved, but noise in images increases
Solution Approach 1:
The imaging process is segmented into multiple sequential low-dose images rather than a single high-dose image. The neural network processes these segmented images individually and combines them to produce a final high-resolution low-noise image, resolving the contradiction between resolution and noise
Solution Approach 2:
A trained neural network serves as an intermediary between the noisy raw images and the final clean image. The network learns to map noisy images to clean images during training, acting as a mediator that removes noise while preserving resolution
2Object-affected harmful factors
If existing machine learning models are used for denoising, then noise removal capability is improved, but requirement for ground truth images creates limitation
Solution Approach 1:
Instead of requiring ground truth images to train the model, the approach inverts the training process by using pairs of noisy images with different noise characteristics. The network learns to identify and remove noise patterns without needing access to the original clean images, making the system adaptable to scenarios where ground truth is unavailable
Solution Approach 2:
The neural network performs self-training by comparing multiple noisy versions of the same image and learning to reconstruct the underlying clean signal from these variations. The system serves itself by using the statistical properties of noise across multiple acquisitions to identify and remove noise without external ground truth data
Data Source
AI summary
In some examples, a machine learning model may be trained to denoise an image. In some examples, the machine learning model may identify noise in an image of a sequence based at least in part, on at least one other image of the sequence. In some examples, the machine learning model may include a recurrent neural network. In some examples, the machine learning model may have a modular architecture including one or more building units. In some examples, the machine learning model may have a multi-branch architecture. In some examples, the noise may be identified and removed from the image by an iterative process.